optimization in container liner shipping
TRANSCRIPT
Optimization in container liner shipping
Judith Mulder1 and Rommert Dekker1
1Econometric Institute, Erasmus University Rotterdam, the Netherlands
Econometric Institute Report 2016-05
Abstract
We will gives an overview of several decision problem encountered in liner shipping.
We will cover problems on the strategic, tactical and operational planning levels as well
as problems that can be considered at two planning levels simultaneously. Furthermore,
we will shortly discuss some related problems in terminals, geographical bottlenecks for
container ships and provide an overview of operations research methods used in liner shipping
problems. Thereafter, the decision problems will be illustrated using a case study for six
Indonesian ports.
1 Introduction
Seaborne shipping is the most important mode of transport in international trade. In comparison
to other modes of freight transport, like truck, aircraft, train and pipeline, ships are preferred for
moving large amounts of cargo over long distances, because shipping is more cost e�cient and
environmentally friendly (Rodrigue et al. 2013). Reviews of maritime transport provided by the
United Nations Conference on Trade And Development (UNCTAD 2014) show that about 80%
of international trade is transported (at least partly) by sea. Sea transport can be separated
into dry bulk (e.g. steel, coal and grain), liquid bulk (e.g. oil and gas) and containerized cargo.
In 2013, containerized cargo is with a total of 1.5 billion tons responsible for over 15% of all
seaborne trade, which resulted in a world container port throughput of more than 650 million
twenty-foot equivalent units (TEUs).
The shipping market comprises three types of operations: tramp shipping, industrial shipping
and liner shipping (Lawrence 1972). Tramp ships have no �xed route, but ensure an immediate
delivery for any type of cargo from any port to any port, resulting in irregular activities. The
behaviour of tramp ships is thus comparable to taxi services. In industrial shipping the cargo
owner also controls the ships used to transport the freight. The objective of industrial operators
is to minimize the cost of shipping the owned cargoes. Liner ships follow a �xed route within a
�xed time schedule and serve many smaller customers. The schedules are usually published online
and demand depends on the operated schedules. Hence, liner shipping services are comparable
to train and bus services.
1
In the next section, we will discuss a variety of liner shipping problems on the strategic, tactical
and operational planning levels, while Section 3 introduces some problems related to terminal
operations. Sections 4 and 5 discuss respectively the in�uence of geographical bottlenecks and
the importance of operations research in solving these problems. This overview is based on the
following overview articles: Ronen (1983, 1993), Christiansen et al. (2004, 2007, 2013), Meng
et al. (2014). In Section 6, a case study is performed for six Indonesian ports to provide insight
into the di�erent problems. Finally, conclusions are drawn in Section 7.
2 Container liner shipping
We will focus on the liner shipping operations concerned with the transport of containers. Liner
shipping operators face a wide variety of decision problems in operating a liner shipping network.
First, at the strategic planning level, the �eet size and mix problem and the market and trade
selection problem need to be solved. In the �eet size and mix problem, operators decide on the
�eet composition and in the market and trade selection problem on which trade route to serve.
At the tactical planning level the network needs to be designed, prices need to be set and empty
containers have to be repositioned. Finally, at the operational level, operators need to determine
the cargo routing through the network and how to deal with disruptions. Furthermore, they can
make adjustments to the earlier set prices and need to determine a plan to store all the containers
on the ship during the loading process. These problems are considered in respectively the cargo
routing, disruption management, revenue management and stowage planning problems. Some
problems have to be considered at both the tactical and the operational planning level, such as
setting the sailing speed and optimizing the bunkering decision and designing a (robust) schedule.
In this section, we will introduce all these decision problems. In these problems we will make use
of the following terminology. Liner shipping operators will also be referred to as liner shipping
companies, liner companies or liners. Liner ships follow �xed routes, which are sequences of port
calls to be made by the ship. Route networks consist of a set of services, which are routes to
which a ship is allocated. Besides publishing their route networks, liner companies also publish
the exact arrival and departure days at each port of call. When we refer to the route together
with the arrival and departure days, we will talk about a schedule. Finally, a round tour refers
to one traversal of a route and a (sea) leg refers to the sailing between two consecutive ports.
2.1 Strategic planning level
The strategic planning level consists of long term decision problems. Generally, these problems
are only solved at most once a year. Examples of long term decision problems in container liner
shipping are: the �eet size and mix problem and the market and trade selection problem.
2.1.1 Fleet size and mix
In the �eet size and mix problem, a liner company decides on how many ships of each type to keep
in its �eet. Container ship sizes have increased substantially because of the growth in container
trade and because of competitive reasons. For example, the Emma Maersk (introduced in 2006)
2
has an estimated capacity of more than 14,500 TEU. Before the introduction of the Emma
Maersk, the capacity of the largest container ship in the world was less than 10,000 TEU. In
2013, Maersk introduced a series of ships belonging to the Triple E class with capacities of over
18,000 TEU, while both MSC and CSCL introduced container ships with a capacity of more than
19,000 TEU in 2015. Ships bene�t from economies of scale when they are sailing at sea, but they
might su�er diseconomies of scale when berthing in ports. However, the e�ect of the economies
of scale at sea is much larger than the e�ect of the diseconomies of scale in ports (Cullinane
and Khanna 1999). Hence, economies of scale in larger container ships can lead to substantial
savings if the capacity of the ship is adequately used. However, if the demand decreases and the
liner company is not able to �ll these large ships any more, higher operational costs are incurred
by these large ships. Therefore, �eet size and mix problems are used to balance the possible
bene�ts from economies of scale with the risk of not being able to use the full capacity of the
ships. Since building a new container ship may take about one year and ships usually have life
expectancies of 25-30 years, future demand and availability of ships play an important role in the
�eet size and mix problem. Pantuso et al. (2014) present an overview of research conducted on
the �eet size and mix problem. Most of these works incorporate ship routing and/or deployment
decisions in order to ensure feasibility of demand satisfaction and capacity constraints.
2.1.2 Market and trade selection
Before a liner container shipping company starts building a network and operating the routes,
it has to decide which trade lanes to participate in. The Asia-Europe trade lane is an example
of a popular trade lane. Clearly, the selected trade lanes in�uence the type and number of ships
required. For example, trade lanes serving the US will usually not use vessels from the Maersk
Triple E class, since they can not sail through the Panama Canal and most ports in the US are
not capable of handling these large vessels. Furthermore, the type and amount of cargoes that
have to be transported and the required sailing frequency may in�uence the ship types used on
the trade lane.
2.2 Tactical planning level
Medium-term decision problems belong to the tactical planning level. Liner companies usually
change their service networks every 6-12 months, but more often in case of worldwide disruptions.
Problems that have to be solved again each time the service network is adjusted are considered to
belong to the tactical planning level. The examples that will be discussed next are: the network
design problem, the pricing problem and the empty container repositioning problem.
2.2.1 Network design
The network design problem in liner shipping can be split into two subproblems. The �rst
subproblem is the routing and scheduling problem, which is concerned with determining which
ports will be visited on each route, in which order the ports will be called at and what the arrival
and departure times at each port will be. Many studies only consider the routing decisions in the
network design problem and do not address the scheduling problem of determining the actual
3
arrival and departure time. The second subproblem considers the �eet deployment and frequency.
Here, the liner company determines which ships will be used to sail each route and with what
frequency the ships will call at the ports along the route. In general, a weekly frequency is
imposed, which facilitates planning by shippers, but this can be relaxed to a biweekly frequency
for low demand routes or multiple port calls per week for high demand routes. Sometimes sailing
speed optimization is considered as a third subproblem of the network design problem, but in
most studies the sailing speed is either assumed to be �xed and known, or will follow directly from
the imposed frequency. Usually, the cargo routing problem is already included (using expected
demand as input) in the network design problem in order to evaluate the pro�tability of a
service network. The cargo routing problem will be discussed in more detail with the operational
planning level problems.
The structure of the routes in a network can be divided into several types, like non-stop services
or end-to-end connections, hub and spoke systems, hub and feeder systems, circular routes,
butter�y routes, pendulum routes and nonsimple routes (?Brouer et al. 2014a). A non-stop
service or end-to-end connection provides a direct connection between two ports: a ship sails
from one port to the other and immediately back to the �rst port; sometimes this is called a
shuttle service, although that also requires a high frequency. In a hub and spoke system, usually
one port is identi�ed as the main or hub port. All other ports (also called feeder ports) are
served using direct services from the hub port. However, it is also possible that multiple hubs
are applied, which are connected with each other and used as transshipment ports to satisfy
demand between di�erent feeder ports, in which case they might also be referred to as main
ports. In the hub and feeder system, feeder ports might also be visited on routes with multiple
port calls. Circular routes are cyclic and visit each port exactly once, while butter�y routes
allow for multiple stops at the same port in one cycle. Pendulum routes visit the same port in
both directions, only in reverse order. Finally, ports can be visited multiple times on nonsimple
routes. Examples of some of these route types are provided in the case study in Section 6.
The liner shipping network design problem has attained quite some attention in the literature.
We will brie�y describe some of the recent publications on this problem. Plum et al. (2014b)
consider a subproblem of the network design problem. They develop a branch-and-cut-and-price
algorithm to �nd a single vessel round trip. Each port has to be visited exactly once and the
best paying demand pairs are accepted and transported. Polat et al. (2014) consider an adapted
neighbourhood search method to solve a hub and feeder system with one single hub. Finally,
Zheng et al. (2015) propose a genetic algorithm to solve the same problem with multiple hubs.
Wang and Meng (2014) propose a column-generation heuristic approach to �nd the best liner
shipping network. Each port can be visited twice during each route: once on the inbound
direction and once on the outbound direction. Brouer et al. (2014a) provide both a base mixed
integer programming formulation for the network design problem and benchmark data instances.
They propose a column generation approach to generate butter�y and pendulum routes. Plum
et al. (2014c) extend the butter�y routes as used in the benchmark model to routes with multiple
butter�y ports. Brouer et al. (2014b) propose a matheuristic to solve the base network design
problem with nonsimple routes. Although their assumptions are a bit more restrictive than in
the benchmark paper (Brouer et al. 2014a), they are able to construct a more pro�table route
network using this approach.
4
Liu et al. (2014), Mulder and Dekker (2014) and Wang et al. (2015) consider slightly di�erent
network design problems. Liu et al. (2014) consider a problem in which the port-to-port demand
is combined with the inland transportation. They start with an initial liner network and try
to improve it while also including the transportation between the ports and the real origin
and destination of the demand. Mulder and Dekker (2014) consider the strategic liner shipping
network design problem, including the �eet size and mix problem, using a hub and feeder network
structure. Wang et al. (2015) consider the liner shipping network alteration problem. In this
problem, an initial liner network is given and this network is modi�ed to become more pro�table.
2.2.2 Pricing
The goal of liner companies is to maximize pro�t by transporting containers from one port to
another. The revenue of the company is determined by the amount of containers that are trans-
ported and the price that will be charged for each container. The pricing problem is concerned
with which price to charge for each possible demand pair. Factors that in�uence the price are
for example: distance, trade direction, expected demand and expected capacity. The pricing
problem is more a marketing, micro-economic problem than an operations research problem. Al-
though it is an interesting problem, it has hardly been touched. Two approaches exist: cost-plus
and what the market can pay. Yet, even determining the cost is a di�cult allocation problem.
2.2.3 Empty container repositioning
Containers delivering import products in a region can be re-used to transport export goods to
another region. However, most regions face an imbalance between import and export containers.
This trade imbalance results in an excess of empty containers in regions with more import than
export and a shortage of containers for high export regions. The empty container repositioning
problem tries to reallocate the empty containers in order to solve the imbalance, where costs are
associated with transporting a container from one region to another. The repositioning of empty
containers is considered to be very costly, since there is no clear revenue associated with it.
Some recent papers dealing with the empty container repositioning problem are the following.
Both Di Francesco et al. (2014) and Long et al. (2015) consider the empty container repositioning
problem under uncertain container demand and use a stochastic optimization approach to solve
it. Zhang and Facanha (2014) consider the problem of repositioning empty containers to the
location of demand in the US. Empty containers are transported using trucks or trains to the
location where they can be loaded. If containers can not be allocated to a loading location, they
are transported to a West Coast port to be shipped to Asia. Huang et al. (2015) consider the
network design problem with both laden and empty container repositioning. Multiple hub ports
are identi�ed, where transshipment from feeder ports might take place. They select the best
routes from a candidate set of routes, which is used as input in the model.
2.3 Operational planning level
The operational planning level captures the problems that occur during the execution of the
routes in the service network. In order to solve operational level problems, reliable information
5
about the actual situation is needed. Hence, operational problems usually need to be solved
relatively shortly before the solutions have to be implemented. Next we will discuss the cargo
routing problem, the disruption management problem, the revenue management problem and
the stowage planning problem.
2.3.1 Cargo routing
The cargo routing problem takes the liner shipping network and container demand as an input.
The goal of this problem is to �nd a cargo �ow over the network, satisfying the capacity con-
straints imposed by the allocated container ships, that maximizes the pro�t of transporting the
containers. Costs are associated to (un)loading and transshipment operations. A transshipment
occurs when a container has to be unloaded from a ship and loaded to another ship in order to
arrive at its destination. Additionally, penalties can be imposed for demand that is not met.
It is also possible to include transit time constraints to guarantee that containers will arrive on
time at their destination.
Formulations of the cargo routing problem can be distinguished in OD-based link �ow formu-
lations, origin/destination-based link �ow formulations, segment-based �ow formulations and
path-based formulations (Meng et al. 2014). All �ow formulations consider the amount of �ow
at a link or segment of the route as decision variables in the model. Flow balance constraints
ensure that all �ow starts at the origin port and arrives at the destination port, but the exact
route followed by a container might not be immediately clear from the model. In the OD-based
link �ow formulation, both the origin and the destination of the container are stored for each link
in the network, while the origin/destination-based link �ow formulations only store the origin
or destination of the container. In this way, the number of decision variables can be reduced
signi�cantly. In a segment-based �ow formulation, consecutive links of a route are already com-
bined into segments before building the model. Segment-based �ow formulations reduce the
number of decision variables even more, but limit the possibility of transshipment operations to
the ports at the beginning and end of the prede�ned segments. Finally, in path-based formu-
lations complete container paths from origin to destination are generated beforehand and used
as decision variable in the model. These paths might also include transshipment operations.
The disadvantage of path-based formulations is that the number of paths might explode, such
that more complex methods, like column generation, are needed to solve the problem. However,
path-based formulations can usually be solved faster than �ow-based formulations. Furthermore,
transit time constraints are easily incorporated in path-based formulations, while this is gener-
ally much more troublesome in �ow-based formulations. Little research is performed on the
separate cargo routing problem: usually it is considered as a subproblem of the network design
problem. Recently, Karsten et al. (2015) considered the cargo routing problem with transit time
constraints. They propose a path-based formulation exploiting the ease to include transit time
in this type of model. Their �ndings show that including transit time constraints in the cargo
routing model is essential to �nd practically acceptable container paths and does not necessary
increase computational times.
6
2.3.2 Disruption management
During the execution of the route schedules, ships may encounter delays. The disruption man-
agement problem focuses on which actions should be taken in order to get back on schedule
after a disruption has occurred. Examples of actions that might be performed are: changing the
sailing speed, swap port calls, cut and go (leave the port before all containers are (un)loaded)
or skip a port. Usually, the goal of disruption management is to �nd a sequence of actions with
minimum cost such that the ship will be back on schedule at a predetermined time. Brouer et al.
(2013) propose a mixed integer programming formulation to solve this problem and prove that
the problem is NP-hard. However, experimental results show that the model is able to solve
standard disruption scenarios within ten seconds to optimality.
2.3.3 Revenue management
At the operational level, more information about the demand and available capacity of a ship is
available. Therefore, it might be pro�table for liner companies to vary their prices based on the
available capacity between a port pair. Liners will probably charge higher prices related to low
capacity pairs, while they might reduce the prices on legs where the available capacity is high.
2.3.4 Stowage planning
The stowage planning problem determines at which location containers are stored on the ship
during the loading process. The stowage planning is a very complicated process with many
constraints. Essential constraints are for example the stability of the ship both during the next
sea leg and during the (un)loading process. Furthermore, containers may have to be stored at
speci�c locations on the ship, like reefer containers. However, the storage of the containers also
in�uence the (un)loading process in the next ports. Ideally, all containers with destination in
the next port are stored on top of the stack, but this may take too many movements in the
current port. Hence, a trade-o� between the number of moves required to store and to discharge
a container has to be made. Tierney et al. (2014) prove that the container stowage planning
problem is a NP-complete problem.
2.4 Both tactical and operational level
Finally, some problems can either be considered at two di�erent planning levels or have to be
considered at two planning levels at the same time. For example, sailing speed optimization
and bunkering optimization can be considered at the moment a new service network is designed,
but the solutions to these problems can be reconsidered during the execution of the routes.
Furthermore, robust schedule design is an example of a decision problem that combines decisions
to be taken at the tactical and at the operational level.
2.4.1 Sailing speed and bunkering optimization
Both sailing speed optimization and bunkering optimization are typical problems that can be
considered at two di�erent planning levels. At the tactical planning level, the environmental
7
aspects of sailing speed are usually considered. At the operational level, sailing speed is mostly
used as an instrument to reduce delays incurred by the ship. The bunkering optimization problem
is concerned with deciding at which ports ships are going to be refuelled. Initially, a bunker
refuelling plan is made given estimates of the bunker price at the moment the ship will be at a
bunkering station. Shipping lines also regularly make bunkering contracts, containing the ports
where bunker can be purchased, the amount to be purchased, the price to be paid and the validity
duration of the contract (Pedrielli et al. 2015). However, due to �uctuations in prices or fuel
consumption, this initial plan might have to be adjusted at the operational planning level. At
this stage, more accurate information about the fuel prices and availability at the ports and the
bunker level of the ship is available. Sailing speed plays an important role in the bunker fuel
consumption of ships and hence sailing speed optimization is often included in the bunkering
optimization problem (Yao et al. 2012).
Recently, the sailing speed and bunkering optimization problems have received increasing atten-
tion. Psaraftis and Kontovas (2014) consider the speed optimization problem at the operational
planning level. They include fuel prices, freight rates, cargo inventory costs and fuel consump-
tion dependencies on payload into their model. Kontovas (2014) and Du et al. (2015) consider
the in�uence of sailing speed on fuel emissions. Wang et al. (2013) provide a literature review
on bunker consumption optimization problems. Bunker consumption is an important input for
bunkering optimization. Yao et al. (2012) study the bunker fuel management strategy for a
single liner shipping route. The strategy consists of the selection of the bunkering ports, the
determination of the bunkering amounts and the adjustments in the sailing speeds. They con-
sider a deterministic situation in which all parameters, including bunker costs, are �xed and
known. Plum et al. (2014a) and Pedrielli et al. (2015) study the problem to determine the op-
timal bunkering contracts. Plum et al. (2014a) propose a mixed integer programming model,
which is solved using column generation. Also, the possibility to purchase bunker on the spot
market is included in their model. Pedrielli et al. (2015) use a game theoretical approach to
design the contracts. Wang et al. (2014) propose a fuzzy approach to include uncertainties in
the bunkering port selection problem. Their method returns a ranking of ports based on the
pro�tability to bunker in these ports. Sheng et al. (2015) implement an (s, S) policy to jointly
optimize the bunkering and speed optimization problem taking into account both bunker price
and consumption uncertainty. Finally, Wang and Meng (2015) consider the robust bunker man-
agement problem, taking into account that the real sailing speed might di�er from the planned
sailing speed.
2.4.2 Robust schedule design
Robust schedule design can be seen as a combination of the scheduling problem at the tactical
level and disruption management or sailing speed optimization at the operational level. The
order in which ports are visited is considered to be an input of this problem. The goal is to
jointly determine the planned arrival and departure times in each port and the actions that will
be performed during the execution of the route when delays are incurred. The di�culty of this
problem is that the tactical and operational planning level problems can not be solved separately,
but have to be considered simultaneously.
8
The problem to determine the scheduled arrival and departure times under uncertainty in port
times and a predetermined sailing speed policy is considered in Qi and Song (2012) and Wang
and Meng (2012). Qi and Song (2012) provide some useful insights in the optimal schedule under
100% service level constraints. Wang and Meng (2012) formulate the problem as a two-stage
stochastic programming problem and solve it using sample average approximation. They are
able to �nd solutions with an objective value within 1.5% of the optimal solution in less than
one hour.
3 Terminal operations
Liner shipping operations are closely related to terminal operations and decisions about ships
cannot be taken while disregarding their e�ects on terminals. In fact, terminals are the largest
bottleneck for shipping. It is important to have the right berth slots and to be loaded and
unloaded quickly and in a predictable manner. There are many ports all over the world with a
large number of ships waiting in front of the harbour to be allowed to berth. Accordingly we will
discuss those terminals operations aspects which directly a�ect shipping, viz. berth scheduling,
crane allocation and container stacking.
3.1 Berth scheduling
Both at a tactical as well as at an operational level, liner shipping schedules are made while
taking berth availability into account. On a tactical level, when designing the liner shipping
routes, agreements are made with terminals on berth availability and productivity (how many
cranes and crane teams will be employed and how many container moves will be done per hour).
This enables the shipping line to calculate the port time of his ships and to complete the ship
scheduling. Naturally bu�er times are incorporated in the schedule and in the berth schedule
in order to take care of schedule deviations. Quite often agreements are made on demurrage
charges (penalties related to delayed cargoes) if terminals need more time or if the shipping line
arrives too late at the terminal. At the operational level the berth schedule is adjusted according
to actual information. Quite often liner ships are too late. In 2015 Drewry shipping reports that
ships are on average one day late. So the berth schedule is updated at a relatively short term
(2 weeks) to take care of changing circumstances, while taking the tactical berth planning as a
start.
3.2 Crane allocation
One level deeper than berth scheduling is the crane allocation. Cranes are used to move a
container from the quay to the location where it will stored on the ship and vice versa. The
storage locations on the ships are called locks. As container ships typically visit many ports,
the cargo destined for a particular port will be distributed over many holds in the ship. After
unloading, a ship will load cargo for several destinations which all have to be put in di�erent
ship holds. As a result the scheduling of the cranes is a di�cult stochastic problem (handling
times of containers are quite variable). The last crane to �nish determines the moment when the
9
ship can leave and hence the port time. A good balancing of the workload between the cranes
is therefore necessary, but also very di�cult to achieve. Another complication comes from the
fact that port workers often work in shifts with �xed starting and end times, and a terminal will
have to accommodate these restrictions.
3.3 Container stacking
A �nal aspect we like to mention is the stacking of the containers in the yard. The issue is not
only that containers are stacked on top of each other, which complicates the retrieval of a bottom
container, it is also the location on the yard of the containers to be loaded. If a ship moors right
before the place where its (to be loaded) containers are located, then travel distances to the
quay cranes are short and no bottlenecks are likely to occur. However, if a ship (due to delay or
congestion at the berths) berths somewhere else, or if containers are spread out over the yard,
then the terminal has to transport the containers over longer distances by which the loading
could potentially be delayed. Container stacking is closely related to stowage planning, as the
latter determines the order in which containers are to be loaded. In a perfect world one can
take the order in which containers are stacked into account while making the load planning, but
that creates a very complex problem, which also su�er from the variations in the loading. Hence
costly reshu�es, where top containers are placed somewhere else to retrieve bottom containers
are needed in large quantities.
4 Geographical bottlenecks
Canal restrictions form the main geographical bottlenecks for container ships. The Suez Canal
and the Panama Canal are two well known canals imposing restrictions on container ships. The
type of restrictions may di�er between di�erent canals. The main restriction imposed by the
Suez Canal is for example the compulsory convoy passage through the canal. This results in long
waiting times if a container ship misses the planned convoy. The Panama Canal on the other
hand, imposes limits on the size of ships that want to sail through the canal.
Two other examples of geographical bottlenecks are the Strait of Malacca and the Gulf of Aden.
These waterways are narrow, but are strategically important locations for the world trade, making
them vulnerable to piracy.
Finally, ports may also impose geographical bottlenecks. Large ships might not be able to access
certain ports, because the access ways have tight draft restrictions.
5 Operations research in liner shipping
In 1983, Ronen provided the �rst overview paper on the contribution of operations research meth-
ods in ship routing and scheduling. Since this �rst paper, every decade a follow-up overview paper
appeared reviewing new research conducted in that decade (Ronen 1993, Christiansen et al. 2004,
2013). Initially, the reviews were mainly focused on the ship routing and scheduling problem, but
more and more other shipping problems are included in these reviews. Furthermore, Christiansen
10
et al. (2007) provides an extensive overview of maritime transportation problems. Finally, Meng
et al. (2014) give an overview of research related to container routing and scheduling in the liner
shipping industry in the last thirty years. The number of citations in these reviews has increased
fast, showing the increasing interest in operations research in liner shipping problems.
Liners usually face complex problems, because the above discussed decision problems cannot be
seen separately from each other and because problem instances are usually large. For example,
when a liner company wants to determine its service network, it has to consider which e�ect
the included routes will have on the cargo routing problem. The solution to the cargo routing
problem depends on the underlying network and will in�uence the pro�t of that network. This
increases the complexity of the problems faced by liners, since they need to solve multiple decision
problems simultaneously. Furthermore, the number of ports that need to be included in a network
is usually large (it can easily contain over 100 ports). The Indonesian case study in the next
section will show that designing a network for only six ports is already quite di�cult. Liner
companies used to solve these problems manually, but in the last years computerized decision
support systems became available. A well-known example of a successful decision support system
is TurboRouter, a tool for optimizing vessel �eet scheduling (Fagerholt and Lindstad 2007).
6 Case study: Indonesia
Shipping is an important mode of transport in Indonesia because the country consists of many
islands. Figure 1 shows six main ports in Indonesia. The six ports are located on �ve di�erent
islands of Indonesia, hence transport over land is only possible between Jakarta and Surabaya.
Transportation between all other combinations of these cities is only possible by sea or air.
Belawan
JakartaSurabaya
Banjarmasin
Makassar
Sorong
Figure 1: Location of six main ports in Indonesia
We will use the Indonesian case to illustrate some of the decision problems introduced in Section 2.
Thereto, we will assume that Table 1 gives the expected weekly demand in TEUs between the
Indonesian ports. The last column and row give the row and column sums, denoting respectively
11
the total supply from and demand to a port. The supply and demand values denote the number
of containers leaving and arriving in the port respectively. The di�erence between demand and
supply indicates how many empty containers have to be repositioned from or to the port. For
the six ports in Indonesia, the empty container repositioning problem is of limited importance,
since there are no large di�erences between supply and demand.
Belawan Jakarta Surabaya Banjarmasin Makassar Sorong Supply
Belawan - 6,500 1,000 100 75 25 7,700
Jakarta 6,750 - 2,000 4,000 2,800 450 16,000
Surabaya 1,000 2,500 - 3,750 4,800 2,150 14,200
Banjarmasin 100 3,600 3,500 - 10 0 7,210
Makassar 100 3,500 4,000 75 - 0 7,675
Sorong 50 650 2,000 0 0 - 2,700
Demand 8,000 16,750 12,500 7,925 7,685 2,625 55,485
Table 1: Expected weekly demand in TEU between the Indonesian ports (Source: own calcula-tions)
Table 1 shows that Jakarta and Surabaya are the two ports with the largest container throughput,
while trade with Sorong is relatively small. In this speci�c case, this might lead to problems, since
Sorong is also located relatively far away from the other ports. Liner shipping companies prefer
to o�er services calling at the ports of Jakarta and Surabaya and consider it too costly to call at
Sorong. By charging higher prices for containers that have to be transported from or to the port
of Sorong, liners can make stops at Sorong more attractive. Hence, the liner company may use
the pricing strategy to ensure that services calling at Sorong will also be bene�cial. However, to
determine exactly which prices they have to charge in order to maximize their pro�t, the liner
company needs more details on the cost structure of the network they will provide.
Figure 2 shows examples of a hub and feeder system, a circular route, a butter�y route and a
pendulum route calling at the six Indonesian ports. In the hub and feeder system of Figure 2a
the port of Surabaya is the hub port, while Belawan, Jakarta, Banjarmasin, Makassar and Sorong
are feeder ports. The route Surabaya - Jakarta - Belawan - Surabaya is referred to as F1. F2 is a
direct feeder route between Surabaya and Sorong. The third feeder route, F3, calls at Surabaya,
Banjarmasin and Makassar after which it returns to Surabaya. The circular route in Figure 2b
has as characteristic that each port is called at exactly once during the round tour. Figure 2c
shows the butter�y route Belawan - Surabaya - Banjarmasin - Makassar - Sorong - Surabaya -
Jakarta - Belawan on which Surabaya is visited twice. Finally, in the pendulum route of Figure
2d all ports are visited twice, only the second time in reversed order.
12
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Soron
g
F1
F3
F2
(a)Example
ofahubandfeeder
system
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Sorong
(b)Example
ofacircularroute
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Soron
g
(c)Example
ofabutter�yroute
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Sorong
(d)Example
ofapendulum
route
Figure
2:Route
examplesfortheIndonesianports
13
Belawan Jakarta Surabaya Banjarmasin Makassar Sorong
Belawan - 1,064 1,488 1,430 1,708 2,807Jakarta 1,064 - 438 614 806 2,102Surabaya 1,488 438 - 328 520 1,816Banjarmasin 1,430 614 328 - 353 1,577Makassar 1,708 806 520 353 - 1,375Sorong 2,807 2,102 1,816 1,577 1,375 -
Table 2: Distances between the Indonesian ports in nmi (Source: www.ports.com/sea-route/)
Table 2 shows the distances in nautical miles (nmi) between the six Indonesian ports and Table 3
provides some characteristics of �ve ship types. Types 1, 2, 3 and 5 are obtained from Brouer
et al. (2014a), while Type 4 is suggested by the Indonesian government and costs are obtained
using interpolation. Note that the fuel usage in ton/day of Type 4 is larger than the usage
of Type 5, because Type 4 has a higher design speed than Type 5. These data can be used
to get some insight in the route cost using di�erent ship types and network structures. In the
calculations we use a simpli�ed version of the fuel cost function as provided in Brouer et al.
(2014a):
Fs(v) = 600 ·(v
v∗s
)3
· fs (1)
Here, Fs(v) denotes the fuel cost in USD per day for a ship of type s sailing at a speed of v knots
(nmi/hour). v∗s and fs are the design speed and fuel consumption in ton per day of a ship of
type s sailing at design speed and can be found in Table 3. Remark that the bunker cost varies
over time, but is assumed to be constant and equal to 600 USD per ton in this study (Brouer
et al. 2014a). Table 4 shows the route distance in nautical miles, the duration in weeks, the
frequency, the number of ships required to obtain the frequency and the sailing speed in knots
for each route. Distances can be found by adding the distances of the individual sea legs, while
the duration and frequency are manually �xed in this example.
Ship Capacity Charter cost Draft Min speed Design speed Max speed Fuel usagetype (TEU) (USD/day) (m) (knots) (knots) (knots) (ton/day)
1 900 5,000 8 10 12 14 18.82 1,600 8,000 9.5 10 14 17 23.73 2,400 11,000 12 12 18 19 52.54 3,500 16,000 12 12 18 20 60.05 4,800 21,000 11 12 16 22 57.4
Table 3: Data of the ship characteristics (Source: Brouer et al. 2014a)
In liner shipping it is common to use weekly port calls at a route. Route durations are typically
an integer number of weeks such that an integer number of ships is needed to sail this route.
For example, a route with duration three weeks and which is sailed by three ships, ensures a
weekly frequency. Given the duration and frequency, the number of required ships can be found
by taking the product of these two values.
14
Route Distance Duration Frequency Required Speed(nmi) (weeks) (per week) ships (knots)
F1 2,990 2 1 2 11.33*
F2 3,632 2 1 2 12.61F3 1,201 1 1 1 12.51Circular 6,476 4 1 4 12.27Butter�y 6,862 4 1 4 13.62Pendulum 7,802 5 1 5 13.55
Table 4: Route characteristics for the di�erent ships (an * indicated that the speed is outsidethe feasible range for some ship types)
10 12 14 16 18 20 220
50
100
150
Speed (knots)
Fuel
cost
(USD/n
m)
Ship 1Ship 2Ship 3Ship 4Ship 5
Figure 3: Fuel cost in USD per nautical mile
Figure 3 shows the fuel price in USD per nautical mile at di�erent speeds for the �ve ship
types. The fuel price is a convex function, meaning that when the speed is doubled, the fuel
cost per nautical mile is more than doubled. Hence, a constant sailing speed during the route
will minimize the fuel cost. The speed is calculated under the assumption that every port call
takes 24 hours and the durations as given in Table 4. The following formula can then be used to
determine the speed on each route:
v =δ
168 · t− 24 · n, (2)
where δ is the route distance in nautical miles, t the route duration in weeks and n the number
of port calls on the route. An ∗ in the column denoting the speed of Table 4 indicates that the
speed is outside the feasible speed range for some ship types. The frequency is chosen in such a
way that it is feasible for each ship type when sailing at maximum speed. Hence, the necessary
speed can only be lower than the minimum speed of the ship type, in which case the ship will
sail at minimum speed and will wait in one of the ports to obtain a weekly frequency.
Table 5 shows the weekly cost in USD for each of the routes given the frequency and duration
as given in Table 4. The route costs consist of three components: the �xed ship costs, the port
15
call costs and the fuel costs. When a liner company needs three ships to satisfy the required
route duration and frequency, it bears weekly the �xed ship costs of all these three ships. Hence,
the �xed ship cost is given by 7 · S · cfs , with S the number of required ships and cfs the daily
�xed ship cost of type s, which can be found in Table 3. The port call cost is the sum of the
port fees of the ports visited on the route. If we assume that all port fees are the same, the port
call cost is given by Fp · n · q, where Fp is the port fee per port visit and q the route frequency.In this example, we assume that Fp = 650 USD. The fuel cost is given by the product of the
frequency, the number of days that a ship needs to sail one round tour and the fuel cost per
day: q · δ24·v ·Fs(v), where Fs(v) is the fuel cost in USD per day when sailing at speed v as given
by (1). Consider a liner route with a duration of two weeks to which four ships are allocated.
Each port on the route will then be called twice a week, resulting in a frequency of twice a week.
Each ship needs two full weeks to sail a round tour, so in one week it will sail half of the route.
Since there are four ships allocated to the route, in total two full round tours are made during
a week (since the frequency is two). This explains the multiplication with the frequency in the
fuel and port call cost. The total route cost in USD per week is now given by:
crs = 7 · S · cfs + q · δ
24 · v· Fs(v) + Fp · n · q, (3)
where crs is the route cost in USD per week for a ship of type s. Doubling the capacity of a ship
will not result in a doubling of the weekly route cost. This illustrates the e�ect of economies of
scale: larger ships will in general have higher total costs, but lower costs per TEU, which is also
exempli�ed in Table 6 by showing the weekly route cost per TEU under the assumption that the
ship is fully utilized. The table also shows that the e�ect of economies of scale can di�er quite a
lot between ship types.
Cost in USD/week
Route/Ship Type 1 Type 2 Type 3 Type 4 Type 5
F1 176,268 196,765 252,848 336,691 446,793F2 228,411 238,026 285,297 373,868 497,669F3 88,076 98,537 121,253 162,296 214,803Circular 408,876 438,257 531,147 702,468 933,207Butter�y 490,525 503,210 598,818 779,713 1,038,189Pendulum 571,488 596,234 714,297 935,318 1,243,643
Table 5: Route cost per week for the duration and frequency as given in Table 4
16
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Soron
g
16,950
17,250
18,000
15,600
14,885
14,875
2,70
0
2,62
5
Cap
acityroutes:
F1:
18,200
F2:
3,50
0
F3:
16,000
(a)Utilizedcapacities
inTEU
forthehubandfeeder
system
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Sorong
27,460
26,710
28,41028,485
28,475
27,760
Cap
acity:28
,800
(b)Utilizedcapacities
inTEU
forthecircularroute
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Sorong
17,025
17,325
18,075
18,225
17,510
17,500
17,575
Cap
acity:19
,200
(c)Utilizedcapacities
inTEU
forthebutter�yroute
Belaw
an
Jakarta
Surabaya
Ban
jarm
asin
Makassar
Sorong
7,70
0
8,00
0
10,450
11,500
11,625
9,86
0
10,310
10,375
2,625
2,70
0
Cap
acity:11
,800
(d)Utilizedcapacities
inTEU
forthependulum
route
Figure
4:Route
exampleswithutilizedcapacities
17
Cost per TEU in USD/week
Route/Ship Type 1 Type 2 Type 3 Type 4 Type 5
F1 196 123 105 96 93F2 254 149 119 107 104F3 98 62 51 46 45Circular 454 274 221 201 194Butter�y 545 315 250 223 216Pendulum 635 373 298 267 259
Table 6: Economies of scale in ship size at full utilization
The disadvantage of the circular route is that the capacity can not be utilized e�ciently. When
containers from for example Surabaya to Jakarta are transported using the circular route, they
will be on board of the ship on all sea legs except the leg from Jakarta to Surabaya. Butter�y
routes are better able to utilize the available capacity, since some ports are visited twice on a
round tour. In the butter�y route, the ports of Surabaya and Jakarta are visited directly after
each other, such that the containers are only on board during one sea leg of the route. The
pendulum route visits all ports twice, hence it needs the lowest capacity. In a hub and feeder
network, usually many ports are connected by only one or a few sea legs. This ensures that hub
and feeder networks are able to utilize the available capacity very e�ciently. Figure 4 shows
the utilized capacity at each sea leg in the four di�erent route networks under the assumption
that all demand has to be satis�ed using only the given network. For the butter�y route, we
assumed that the containers that have to be transported from Makassar to Banjarmasin will
stay on board of the ship during the route segment Surabaya - Jakarta - Belawan - Jakarta
- Surabaya. Alternatively, these containers can be unloaded during the �rst call at Surabaya
and loaded again during the second port call at Surabaya in which case transshipment costs at
Surabaya are incurred. The utilized capacities are found by adding all container �ows that need
to traverse the given sea leg in order to arrive at their destination. Table 7 shows the required
capacity in TEU for each route, the number of port calls per week for each ship type in order
to have enough capacity to satisfy all demand, the available capacity in TEU using these ship
types and the total route costs in USD per week. The required capacity is found by taking the
maximum utilized capacity of the route. Next, we make a combination of ship types such that
enough capacity is available at each route. Given these ship allocations, the total route cost can
be found by multiplying the weekly route cost for a ship type by the number of port calls per
week divided by the route frequency. Note that the type and number of ships needed vary a lot
between the three di�erent route structures. For the hub and feeder system, 2 ·1+1 ·1 = 3 ships
of type 2 (since feeder route 1 has a duration of 2 weeks and feeder route 3 has a duration of
one week), 2 · 2 + 2 · 1 = 6 ships of type 4 and 2 · 2 + 1 · 3 = 7 ships of type 5 are needed. The
circular route uses 4 · 6 = 24 ships of type 5, while the butter�y route uses 4 · 4 = 16 ships of
type 5. Finally, the pendulum route uses 5 · 2 = 10 ships of type 4 and 5 · 1 = 5 ships of type 5.
Hence, the optimal solution to the �eet size and mix problem is highly dependent on the network
structure.
Table 7 also shows the total network cost for the hub and feeder system, the circular route, the
18
Route Req. cap. Port calls per week Av. cap. Cost
(TEU) Type 1 Type 2 Type 3 Type 4 Type 5 (TEU) (USD/week)
F1 18,000 0 1 0 2 2 18,200 1,763,733F2 2,700 0 0 0 1 0 3,500 373,868F3 15,600 0 1 0 0 3 16,000 742,948HF-Total 2,880,549
Circular 28,485 0 0 0 0 6 28,800 5,599,239
Butter�y 18,225 0 0 0 0 4 19,200 4,152,755
Pendulum 11,625 0 0 0 2 1 11,800 3,114,280
Table 7: Network cost per week when shipping all demand
butter�y route and the pendulum route. The table indicates that the hub and feeder system
and the pendulum route are by far the cheapest choices of networks in this example. They
both cost approximately 3 million USD per week, while the circular and butter�y routes cost
respectively about 5.5 and 4 million USD per week. One remark has to be made: in the hub and
feeder system, a lot of containers need to be transshipped, adding additional costs that are not
included in this example. In total 15,450 containers have to be transshipped per week in the hub
and feeder system. If a transshipment costs for example 100 USD per container, the total cost
of the hub and feeder system will rise to almost 4.5 million USD per week. Hence, the hub and
feeder system will then have higher costs than the butter�y and pendulum routes. Of course,
one could also make route networks with combinations of these routes, which might be more cost
e�cient.
The good performance of the hub and feeder system and pendulum route is (partly) caused
because of the better utilization of capacity in the hub and feeder system. Another advantage
of hub and feeder systems is that liners can allocate di�erent ship types to the di�erent types of
routes. Feeder ports usually have less demand than hub ports, hence it makes sense to allocate
smaller ships to the feeder routes than to the main routes. If all ports are visited on similar
routes, like circular, butter�y and pendulum routes, all these ports are visited by the same ship
type. Hence, large ships might visit very low demand ports if these ships are able to berth in the
smaller ports (smaller ports might have stricter draft restrictions than hub ports). Otherwise
many small ships are needed in order to satisfy the demand of the large ports. However, a
disadvantage of hub and feeder networks is that usually many transshipments are needed in
order to satisfy the demand, which increases both the transportation price and transit time. In
airline passenger transport, hub and feeder systems are very popular; an important reason for
this is that transshipments are made by passengers at no apparent cost.
Finally, we determine the pro�t and e�ciency of the networks when we assume that each con-
tainer will generate a revenue of 200 USD if it is transported, (un)loading and transshipment
costs are all 40 USD per container. Recently, the problem is also studied on request of the In-
donesian government, resulting in a single pendulum route to be sailed. This pendulum route is
also known under the name Pendulum Nusantara. We use the mixed integer programming model
proposed in Mulder and Dekker (2016) to determine the optimal route network given an initial
set of routes. Routes are constructed in the following way using the ordering of ports used for
19
Belawan
JakartaSurabaya
Banjarmasin
Makassar
Sorong
Capacity: 3,500
(a) Route 1
Belawan
JakartaSurabaya
Banjarmasin
Makassar
Sorong
Capacity: 3,50
(b) Route 2
Belawan
JakartaSurabaya
Banjarmasin
Makassar
Sorong
Capacity: 4,800
(c) Route 3
Figure 5: Optimal route network
20
the pendulum route. Ports may be visited at most twice during a route: once on the eastbound
trip and once on the westbound trip. All feasible routes are generated and given as input to the
mixed integer programming problem, which makes use of a path formulation to solve the cargo
allocation. Table 8 shows the pro�ts of these networks and Figure 5 shows optimal route net-
work. We see that the pendulum route performs indeed better than the hub-and-feeder network
and the butter�y and circular routes. The e�ciency of the networks is measured by the shipped
distance in nmi per TEU. The shipped distance for direct shipping is equal to 836.57 nmi/TEU.
Table 8 shows that the pendulum and optimal networks are both e�cient networks with respect
to shipped distance. Furthermore, the hub-and-feeder network is much more e�cient than the
circular and butter�y routes as expected.
The optimal route network as shown in Figure 5 consists of two pendulum routes (Routes 2 and
3) and a non-stop service (Route 1), which is a special type of pendulum route with only two
port calls. The pendulum route structure ensures e�cient transportation between all demand
pairs. All routes have a frequency of once a week, reducing the number of required ships.
Network Shipped distance (nmi/TEU) Pro�t (USD)
Hub-and-feeder 1,428.06 3,159,651Circular 3,269.79 1,058,961Butter�y 2,227.57 2,284,445Pendulum 996.80 3,642,916Optimal 852.66 4,897,109
Table 8: E�ciency and pro�t of the di�erent networks
7 Conclusion
Maritime transportation is very important in the world economy. The types of operations in
the shipping market are distinguished in tramp, industrial and liner shipping. This chapter
considers decision problems that occur in the operations of container liner shipping companies.
The decision problems can be distinguished in three di�erent planning levels: strategic, tactical
and operational. The strategic planning level consists of long term decision problems, while the
tactical and operational planning levels contain respectively medium and short term problems.
This chapter discusses the �eet size and mix and market trade selection problem on the strategic
level, network design, pricing and empty container repositioning on the tactical level and cargo
routing, disruption management, revenue management and stowage planning on the operational
planning level. Furthermore, sailing speed and bunkering optimization and robust schedule
design are covered. These decision problems are related to both the tactical and the operational
problem. The chapter also introduces three decision problems in the terminal operations that
are related to container liner shipping: berth scheduling, crane allocation and container stacking.
A case study is used to explain the concepts of the problems in more detail. The case study is
based on six main ports in Indonesia. Three networks with di�erent structures (hub and feeder
system, pendulum route and butter�y route) are proposed. Calculations show that the cost
21
of the hub and feeder system is much smaller than the cost of the other two networks when
transshipment costs are not considered. Explanations for this result are that capacity is better
utilized in hub and feeder systems compared to the other network structures and ships can be
chosen more freely, since more shorter routes are used. When transshipments are charged at 100
USD per container, the hub and feeder system performs comparable to the butter�y route and
still considerably better than the pendulum route.
References
Brouer, B.D., J.F. Álvarez, C.E.M. Plum, D. Pisinger, M.M. Sigurd. 2014a. A Base Integer Programming
Model and Benchmark Suite for Liner-Shipping Network Design. Transportation Science 48(2)
281�312.
Brouer, B.D., G. Desaulniers, D. Pisinger. 2014b. A matheuristic for the liner shipping network design
problem. Transportation Research Part E 72 42�59.
Brouer, B.D., J. Dirksen, D. Pisinger, C.E.M. Plum, B. Vaaben. 2013. The Vessel Schedule Recovery
Problem (VSRP) - a MIP model for handling disruptions in liner shipping. European Journal of
Operational Research 224(2) 362�374.
Christiansen, M., K. Fagerholt, B. Nygreen, D. Ronen. 2007. Maritime Transportation. C. Barnhart,
G. Laporte, eds., Handbook in OR & MS , vol. 14. Elsevier B.V., 189�284.
Christiansen, M., K. Fagerholt, B. Nygreen, D. Ronen. 2013. Ship routing and scheduling in the new
millennium. European Journal of Operational Research 228(3) 467�483.
Christiansen, M., K. Fagerholt, D. Ronen. 2004. Ship Routing and Scheduling: Status and Perspectives.
Transportation Science 38(1) 1�18.
Cullinane, K., M. Khanna. 1999. Economies of Scale in Large Container Ships. Journal of Transport
Economics and Policy 33(2) 185�207.
Di Francesco, M., M. Lai, P. Zuddas. 2014. A multi-scenario model for empty container repositioning
with uncertain demand. International Journal of Services and Operations Management 19(2)
212�228.
Du, Y., Q. Meng, Y. Wang. 2015. Budgeting the fuel consumption of a container ship over a round
voyage via robust optimization. Transportation Research Record: Journal of the Transportation
Research Board 2477 68�75.
Fagerholt, K., H. Lindstad. 2007. Turborouter: An Interactive Optimisation-Based Decision Support
System for Ship Routing and Scheduling. Maritime Economics & Logistics 9 214�233.
Huang, Y.F. J.K. Hu, B. Yang. 2015. Liner service network design and �eet deployment with empty
container repositioning. Computers & Industrial Engineering 89 116�124.
Karsten, C.V., D. Pisinger, S. Ropke, B.D. Brouer. 2015. The time constrained multi-commodity network
�ow problem and its application to liner shipping network design. Transportation Research Part
E 76 122�138.
Kontovas, C.A. 2014. The Green Ship Routing and Scheduling Problem (GSRSP): A conceptual ap-
proach. Transportation Research Part D 31 61�69.
Lawrence, S.A. 1972. International Sea Transport: The Years Ahead.
Liu, Z., Q. Meng, S. Wang, Z. Sun. 2014. Global intermodal liner shipping network design. Transportation
Research Part E 61 28�39.
22
Long, Y., E.P. Chew, L.H. Lee. 2015. Sample average approximation under non-i.i.d. sampling for
stochastic empty container repositioning problem. OR Spectrum 37 389�405.
Meng, Q., S. Wang, H. Andersson, K. Thun. 2014. Containership Routing and Scheduling in Liner
Shipping: Overview and Future Research Directions. Transportation Science 48(2) 265�280.
Mulder, J., R. Dekker. 2014. Methods for strategic liner shipping network design. European Journal of
Operational Research 235(2) 367�377.
Mulder, J., R. Dekker. 2016. Will liner ships make fewe port calls per route? Econometric Institute
Report EI2016-04, Erasmus University Rotterdam.
Pantuso, G., K. Fagerholt, L.M. Hvattum. 2014. A survey on maritime �eet size and mix problems.
European Journal of Operational Research 235(2) 341�349.
Pedrielli, G., L.H. Lee, S.H. Ng. 2015. Optimal bunkering contract in a buyer-seller supply chain under
price and consumption uncertainty. Transportation Research Part E 77 77�94.
Plum, C.E.M., P.N. Jensen, D. Pisinger. 2014a. Bunker purchasing with contracts. Maritime Economics
& Logistics 16(4) 418�435.
Plum, C.E.M., D. Pisinger, J.J. Salazar-González, M.M. Sigurd. 2014b. Single liner shipping service
design. Computers & Operations Research 45 1�6.
Plum, C.E.M., D. Pisinger, M.M. Sigurd. 2014c. A service �ow model for the liner shipping network
design problem. European Journal of Operational Research 235(2) 378�386.
Polat, O., H.O. Günther, O. Kulak. 2014. The feeder network design problem: Application to container
services in the Black Sea region. Maritime Economics & Logistics 16(3) 343�369.
Psaraftis, H.N., C.A. Kontovas. 2014. Ship speed optimization: Concepts, models and combined speed-
routing scenarios. Transportation Research Part C 44 52�69.
Qi, X., D.P. Song. 2012. Minimizing fuel emissions by optimizing vessel schedules in liner shipping with
uncertain port times. Transportation Research Part E 48(4) 863�880.
Rodrigue, J.P., C. Claude, B.Slack. 2013. The geography of transport systems. 3rd ed. Routledge.
Ronen, D. 1983. Cargo ships routing and scheduling: Survey of models and problems. European Journal
of Operational Research 12(2) 119�126.
Ronen, D. 1993. Ship scheduling: The last decade. European Journal of Operational Research 71(3)
325�333.
Sheng, X., E.P. Chew, L.H. Lee. 2015. (s,S) policy model for liner shipping refueling and sailing speed
optimization problem. Transportation Research Part E 76 76�92.
Tierney, K., D. Pacino, R.M. Jensen. 2014. On the complexity of container stowage planning problems.
Discrete Applied Mathematics 169 225�230.
UNCTAD. 2014. Review of Maritime Transport 2014. United Nations Conference on Trade And Devel-
opment.
Wang, S., Z. Liu, Q. Meng. 2015. Segment-based alteration for container liner shipping network design.
Transportation Research Part B 72 128�145.
Wang, S., Q. Meng. 2012. Robust schedule design for liner shipping services. Transportation Research
Part E 48(6) 1093�1106.
Wang, S., Q. Meng. 2014. Liner shipping network design with deadlines. Computers & Operations
Research 41 140�149.
Wang, S., Q. Meng. 2015. Robust bunker management for liner shipping services. European Journal of
Operational Research 243(3) 789�797.
23
Wang, S., Q. Meng, Z. Liu. 2013. Bunker consumption optimization methods in shipping: A critical
review and extensions. Transportation Research Part E 53 49�62.
Wang, Y., G.T. Yeo, A.K.Y. Ng. 2014. Choosing optimal bunkering ports for liner shipping companies:
A hybrid Fuzzy-Delphi-TOPSIS approach. Transport Policy 35 358�365.
Yao, Z., S.H. Ng, L.H. Lee. 2012. A study on bunker fuel management for the shipping liner services.
Computers & Operations Research 39(5) 1160�1172.
Zhang, Y., C. Facanha. 2014. Strategic planning of empty container repositioning in the transpaci�c
market: a case study. International Journal of Logistics Research and Applications 17(5) 420�439.
Zheng, J., Q. Meng, Z. Sun. 2015. Liner hub-and-spoke shipping network design. Transportation Research
Part E 75 32�48.
24